The "preferred" buffer size for raw_ostream is set to BUFSIZ which on
Windows is only 512. This results in more calls to write and this
overhead can have a significant negative impact on performance,
especially when Anti-Virus is also involved.
Therefore increase the "preferred" buffer size to 16KB for Windows.
One example of where this helps is the LLD --Map option which dumps out
the symbol map for a link. In a link of UE4, this change has been seen
to improve the performance of the symbol map writing by more than a
factor of 6.
Differential Revision: https://reviews.llvm.org/D147340
All the IEEE formats are quite similar, we can merge their code
effectively by writing it parametrically via the fltSemantics object.
We can metaprogram the implementation such that this parametricity is
zero-cost.
New versions I2.1, F2.2, D2.2 A2.1
Make F and Zfinx imply Zicsr.
Make G imply Zifencei.
This should have no impact to generated code. We have no plans to require Zicsr/Zifencei extension to be explicitly enabled to use Zicsr/Zifencei instructions in assembly.
See https://reviews.llvm.org/D147183 for documentation regarding what version specification we implement.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D147179
This patch adds the basic MC layer support for Zicond, based on
[1.0-rc1](https://github.com/riscv/riscv-zicond/releases/tag/v1.0-rc1).
As with other extensions, if there are additional changes between
release candidates without incrementing the version number we won't be
able to reflect that in the version number. I believe we've previously
decided this is not a problem for extensions still considered
experimental (i.e. not yet ratified).
Differential Revision: https://reviews.llvm.org/D146946
Tools such as llvm-objdump will currently inputs when the base ISA has
an unrecognised version. I addressed a similar issue in LLD in D144353,
introducing parseArchStringNormalized. While it would make sense to
migrate `llvm/lib/Object/ELFObjectFile.cpp` to using
`parseArchStringNormalized` as well, this patch takes a less ambitious
initial step. By tweaking the behaviour of `parseArchString` when
`IgnoreUnknown` is true (which only has one in-tree user), we use the
default supported ISA version when a base ISA with unrecognised version
is encountered.
This means that llvm-objdump and related tools will function better for
objects produced from a recent GCC. This isn't a full fix, as
IgnoreUnknown means that an imafd object with attributes specifying
newer A/F/D versions will have those extensions ignored.
Differential Revision: https://reviews.llvm.org/D146070
parseNormalizedArchString adds a code path that creates a RISCVISAInfo
including extensions that may not be supported by LLVM (rather than
erroring or just ignoring them). Therefore, toFeatureVector needs to
check the extension is supported in order to avoid creating unrecognised
feature strings.
This change shouldn't impact any code paths used outside of test code,
but this will be relied upon by the next patch which moves llvm-objdump
and related tools over to using parseNormalizedArchString.
Differential Revision: https://reviews.llvm.org/D146113
X. Sun et al. (https://dl.acm.org/doi/10.5555/3454287.3454728) published
a paper showing that an FP format with 4 bits of exponent, 3 bits of
significand and an exponent bias of 11 would work quite well for ML
applications.
Google hardware supports a variant of this format where 0x80 is used to
represent NaN, as in the Float8E4M3FNUZ format. Just like the
Float8E4M3FNUZ format, this format does not support -0 and values which
would map to it will become +0.
This format is proposed for inclusion in OpenXLA's StableHLO dialect: https://github.com/openxla/stablehlo/pull/1308
As part of inclusion in that dialect, APFloat needs to know how to
handle this format.
Differential Revision: https://reviews.llvm.org/D146441
RV32E was recently [ratified](afd613691c) so we should update the version as our MC-layer support is complete.
Reviewed By: kito-cheng
Differential Revision: https://reviews.llvm.org/D144384
Instead of having a separate single letter and multiletter ranking
use a unified rank that assigns multiletter a larger value than
single letter.
Once we've ranked the extensions, then we compare using these ranks.
Reviewed By: kito-cheng
Differential Revision: https://reviews.llvm.org/D146273
This field is never used in the compiler and was only used in
unit tests added recently.
It's only used as the value in a map where the extension name
is the key. If we need the string we can get it from the key.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D146197
RISCVISAInfo::parseArchString would sometimes error for arch strings
with a trailing _ and sometimes accept them. This patch makes it
consistently error.
Differential Revision: https://reviews.llvm.org/D145949
There is no versioning scheme for the 'g' shorthand for imafd (or in
current ISA specs, imafd_zifencei_zicsr). As such, the only sensible
behaviour to me seems to be to reject a version for it.
Differential Revision: https://reviews.llvm.org/D145954
Empty StringRefs are usually identified by their length being zero, and
sometimes they'll have Data==nullptr (e.g. default constructed, or derived from
an operation like split/copy and result turned out to be empty).
If such StringRef objects are passed to llvm::MemoryBuffer::getMemBufferCopy,
it'll result in UB as neither src nor dst can be null, even if size is zero.
This patch prevents that UB by not issuing a copy whenever StringRef is empty.
Differential Revision: https://reviews.llvm.org/D144706
Prior to this patch, unrecognised z/s/sx/x prefixed extensions were not
ignored when IgnoreUnknown=true.
The first version of this patch, a7313f83b9ca9, incorrectly used
`!isSupportedExtension(Ext)` rather than `!isSupportedExtension(Name)`.
i.e. checked the full substring rather than the split out name, causing
incorrect behaviour when a version is specified. This was fixed and a
new test case addded.
Differential Revision: https://reviews.llvm.org/D145882
Prior to this patch, unrecognised z/s/sx/x prefixed extensions were not
ignored when IgnoreUnknown=true.
Differential Revision: https://reviews.llvm.org/D145882
I believe this implies relationship is documented in the current
spec but may have been less clear in an older spec.
We used to report an error so I think it should be ok to change this.
Only someone expecting the error should be impacted.
Reviewed By: asb, luismarques
Differential Revision: https://reviews.llvm.org/D145125
To have line number information in stack traces (per stack frame) it's
necessary to implement findModulesAndOffsets.
Differential Revision: https://reviews.llvm.org/D142282
This patch adds AArch64 CodeGen support such that the type can be passed
and returned to/from functions, and also adds support to use this type in
load/store operations and PHI nodes.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D136862
To have line number information in stack traces (per stack frame) it's
necessary to implement findModulesAndOffsets.
Differential Revision: https://reviews.llvm.org/D142282
This patch follows on from this RFC thread
<https://discourse.llvm.org/t/rfc-resolving-issues-related-to-extension-versioning-in-risc-v/68472/>
and the ensuing discussion in the RISC-V LLVM sync-up call. The
consensus agreed that the behaviour change in LLD introduced in D138550
that results in object files including arch attributes with unrecognised
extensions or versions of extensions is a regression and should be
treated as such. As it stands, this logic means that LLD will error out
if trying to link a RISC-V object file from LLVM HEAD (rv32i2p0/rv64i2p0)
with one from current GCC (rv32i2p1/rv64i2p1 by default).
There's a bigger discussion about exactly when to warn vs error and so
on, and how to control that, and this patch doesn't attempt to address
all those questions. It simply tries to fix the problem with a minimally
invasive change, intended to be cherry-picked for 16.0.x (ideally
16.0.0, but queued for 16.0.1 if we're too late in the release cycle).
As you can see from the test changes, although the changed logic is
mostly more permissive, it will reject some embedded arch strings that
were accepted before. Because the same logic was previously used for
parsing human-written -march as for the arch attribute (intended to be
stored in normalised form), various short-hand formats were previously
accepted. Maintaining support for such ill-formed attributes would
substantially complicate the logic, and given the previous
implementation was so much stricter in every other way, was surely a bug
rather than a design choice.
Surprisingly, the precise rules for how numbers can be embedded into
extension names isn't fully defined (there is a PR to the ISA manual
that is not yet merged
<https://github.com/riscv/riscv-isa-manual/pull/718>).
In the absence of specific criteria for rejecting extensions names that
would be ambiguous in abbreviated form,
`RISCVISAInfo::parseArchStringNormalized` just pulls out the version
information from the end of each extension description.
Differential Revision: https://reviews.llvm.org/D144353
This gives our worker threads some names that allow easily identifying them in tools such as profilers and debuggers.
Differential Revision: https://reviews.llvm.org/D144297
This fixes the following warning on Windows with latest Clang:
```
[160/3057] Building CXX object lib/Support/CMakeFiles/LLVMSupport.dir/Signals.cpp.obj
In file included from C:/git/llvm-project/llvm/lib/Support/Signals.cpp:260:
C:/git/llvm-project/llvm/lib/Support/Windows/Signals.inc(834,15): warning: comparison of integers of different signs: 'int' and 'unsigned int' [-Wsign-compare]
if (RetCode == (0xE0000000 | EX_IOERR))
~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~
1 warning generated.```
